The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the subprocess module in run_eval.py, run_loop.py, and improve_description.py to execute the claude CLI and local Python scripts. It is explicitly authorized to run Python commands via the allowed-tools frontmatter field. These capabilities are used exclusively for skill auditing and evaluation.
[EXTERNAL_DOWNLOADS]: The eval_review.html template references Google Fonts. This is a reference to a well-known service for UI purposes and is documented neutrally as a safe reference.
[SAFE]: (Indirect Prompt Injection Surface) The skill processes external skill content for auditing and evaluation. While this creates a potential surface for indirect prompt injection if an audited skill contains malicious instructions, the tool is intended for developer workflows and uses structured data handling for its reporting.
[SAFE]: (Dynamic Execution) The skill constructs temporary command files and executes them to verify behavior. These operations are constrained to the .claude/commands/ directory and include automatic cleanup in finally blocks to prevent unintended persistence.

skill-audit